llms-txt-score

Score an llms.txt file against the public awesome-llms-txt rubric. Like Lighthouse, but for llms.txt.

Install

npx llms-txt-score https://stripe.com/llms.txt

Or globally:

npm install -g llms-txt-score
llms-txt-score https://resend.com/llms.txt

Usage

llms-txt-score <url-or-file> [options]

  --format=json|markdown   output format (default: json)
  --out=<path>             write to a file
  --check-links            sample 8 links and HEAD them for stability points
  --content-type=<ct>      override Content-Type when scoring a local file
  --no-color               disable ANSI color in markdown output

What it scores (100 points)

| Weight | Criterion | |------:|---| | 18 | Spec compliance: H1, blockquote, well-formed [name](url): note lists | | 20 | Coverage: section count, total links, canonical-section diversity | | 14 | Agent-action declarations: llms-full.txt, .md twins, MCP, OpenAPI | | 10 | Linked-content stability: sampled HEAD checks | | 10 | Freshness: Last-Modified within 90 days | | 8 | Discoverability: served at /llms.txt, no redirect chain | | 8 | Auth signposting: keywords + dedicated auth section | | 6 | Size discipline: context-window friendliness | | 4 | Content-Type & encoding: text/markdown; charset=utf-8 | | 2 | Voice: plain language, no marketing fluff |

Full rubric, weights, and rationale: RUBRIC.md.

Grade bands (Mozilla Observatory style, capped at 100): A+ ≥ 95, A ≥ 85, A- ≥ 80, B+ ≥ 75, B ≥ 65, C ≥ 50, D ≥ 35, F < 35.

Output

--format=json (default) returns a stable shape, safe to commit as score.json:

{
  "schema_version": 1,
  "score": 87,
  "grade": "A",
  "components": [
    { "id": "spec_compliance", "points": 17, "max": 18, "reasons": [] },
    ...
  ],
  "parsed": { "h1": "Stripe Documentation", "section_count": 23, "link_count": 472 }
}

--format=markdown is meant for humans / CI logs.

Why no dependencies

Zero runtime deps. The CLI is node ./bin/llms-txt-score.js and works with any Node ≥ 18. Anyone can clone the repo, run the tool, and get the same number we did.

License

MIT. See LICENSE.